使用了<a target="_blank" rel="ugc" href="https://www.aplaybox.com/details/model/HmB8k430e520">【崩坏：星穹铁道】昔涟 流云景</a> 3D模型渲染图片和基于<a target="_blank" rel="ugc" href="https://civarchive.com/models/1550626/cyrene-honkai-star-rail-oror"> cyrene (honkai star rail) || 昔涟 崩坏：星穹铁道 </a>生成的图像混合训练得到了一个更好的IL基底模型的Lora，脸更像昔涟而不是爱莉希雅，能够正确展示昔涟的正式着装，但极端情况下眼睛会绘制错误。Using 3D model-rendered images of <a target="_blank" rel="ugc" href="https://www.aplaybox.com/details/model/HmB8k430e520">【崩坏：星穹铁道】昔涟 流云景</a> and images generated based on <a target="_blank" rel="ugc" href="https://civarchive.com/models/1550626/cyrene-honkai-star-rail-oror">cyrene (honkai star rail) || 昔涟 崩坏：星穹铁道 </a>to train a better LoRA model based on IL. It can correctly display Cyrene's face, instead of Elysia's one, and official clothes, but in extreme cases, the eyes may be drawn incorrectly.<hr />v1.1 IL 更新通过删除了错误的标注和重新裁剪了部分空白过多的图片，缓解眼睛绘制模糊的问题，增强了全方位表现能力，无论是从正面还是从背后都有较强的表现能力和一致性，但有时会出现3d渲染风格比较明显的问题By removing incorrect annotations and recropping images with excessive blank space, the depiction of eyes has been enhanced, alleviating the issue of occasionally blurry eye rendering. It now boasts stronger all-around performance capabilities, delivering more powerful and consistent representation whether viewed from the front or the back. However, sometimes you may generate a image with 3d renedering style, just generate another image.<hr />v1.1 NAI在NAI3上训练了一个32维的lora，色彩更好，但一致性更差32 dim lora trained on NAI3, with better color but worse consistencr

Better Cyrene (Honkai: Star Rail) || 更好的昔涟 崩坏：星穹铁道

使用同样的配置在NAI3基底模型上训练的lora，色彩更鲜艳漂亮，但眼睛画得不是很正确，细节把握可能弱于IL版本（注：未开启V预测训练）

Files with SHA256: 77df1b9fdb58a35f11db6e14bfcece2efe4ed376eabb4a86dcb64cbacc52f8b7